Data, Acquisitions, and AI: Insights from FiscalNote's CTO

Vlad Eidelman, Chief Technology Officer at FiscalNote a Government Relationship Management (GRM) service, sits down with David Joy to examine the technical complexities involved with acquisitions.

Vlad explains the challenges with integrating products running on different infrastructures, and talks through his proven strategies for a successful integration.

Join as we discuss:

• The intricacies of tech acquisitions including the essential steps in pre-acquisition diligence, technology integration, and organizational alignment.

• Considerations for creating reliable AI solutions.

• How big data has shifted over the last decade.

David Joy:

What is up, everyone? And thanks for tuning in. In today’s episode of the Big Ideas in App Architecture podcast, we speak to Vlad Eidelman, who is the CTO at FiscalNote. Vlad and I talk about FiscalNote, its early startup dates to their core identity around a data and intelligence company. We also get into his journey as a computer scientist to growing into the role of a CTO and other different aspects around building a startup product, their technology, generative AI, and much more. So bump up that volume and get ready for an intriguing conversation with Vlad Eidelman. Really excited to talk to you. I’ve been doing a little bit of stalking about on you. I went and saw a photo of you at the New York Stock Exchange with FiscalNote going IPO, found that you have a website called Machine Opining. And I did a little bit of prep here, so hopefully I can make sure that it’s worth the time. All right, awesome. So tell me a little bit about you and what you do at FiscalNote, not just for me, for the people who are listening to you right now.

Vlad Eidelman:

I came here to the US in the early ’90s. I think I’ve always been interested in how machines, which have been coming online basically since I was in elementary school, middle school, how machines, how computers can really help with our day-to-day lives of what we want to do. At first, obviously, it was very early playing games on floppy disks that quickly evolved into starting to learn how to code basically in middle school, taking the coding classes in high school. And then I really got a passion for computer science. So you might find out a little bit later in this discussion, I like arguing a lot. I like basically testing ideas against each other so our ideas can fail much quicker than we can in real life and I think that helps validate a lot of the work that we want to do or could do. And so I knew I wanted to have that side of me, so that was more aligned with philosophy. I can reference that back now.

So there was a part of me that really wanted to get into philosophy and understand how the world works, epistemology, metaphysics, and understand how we fit in. And then the other side of me that was already there that, I guess, I can give name to now, again, is the builder side, is to try to create automation using computers, was really interested in how to marry those two. And so in college and undergrad, I was doing comsci and philosophy. And the natural consequence of that was getting into the AI courses that we were offering and starting to do machine learning. And in my undergrad, I was really focused on natural language processing, so working on unstructured data, but specifically the language interactions that we have. And this was early days before IBM Watson before a lot of the things that you’re now accustomed to seeing.

And so people had a lot of really grandiose visions for what we could do, but not quite there in terms of the technology. And so I was really interested in pursuing that, so I got my PhD in computer science specifically in developing large-scale systems for language processing using machine learning and various other things. And then when I was coming out of that, I was then, again, interested in this entrepreneurial mindset of how can I learn as much as possible about what it means to apply this in a real-world setting. So often, the choices are, and this is not true anymore, but at the time, going into a research lab and being a part of either academic or industry research lab, and that’s obviously exciting and a lot of friends and coworkers in those places, but I was really interested in joining really early-stage startup where I could practice my skillset so I wouldn’t ignore that, right? It would still be a core part of the necessary needs, but also learn how everything else works, how to basically apply that in a way that builds great products, helps people understand and do things.

And so I’m around people who are lawyers, and so I live in DC, so that’s, again, a natural consequence of the area I’m in. And so a lot of my interest was in computational law, computational journalism, and how do those things evolve. Basically what we’re seeing now with LLMs, there’s so much content creation that can happen automatically, where is that right marriage of human expertise and automation that can build that in. And so FiscalNote, when I joined a little over 10 years ago, it was fewer than 10 people. So I joined as one of the first employees. And then I’ve been lucky enough to be part of that ride, like you said, from there to when we went in public on the New York Stock Exchange in ‘22. And so we’ve been public for a few years now. I can compare the entire cycle of it. But essentially, what FiscalNote does is we take a lot of both public and bespoke content that we’re creating about geopolitical, policy, market intelligence and use that to power a number of different SaaS applications as well as data products.

And so whether you’re a small nonprofit or a huge F100, F500 multinational organization, you have risks and opportunities associated with different people in government with different agencies, with different governments across the world are doing, and so we try to collect all that information from the outside world and build analysis using AI, that’s where I come in and my teams come in, and the actual content specialist and the subject matter expertise in different areas of the world and different industries marry those two to provide that insight to our clients who then can either act by advocating for or against certain policies or adopting that internally. And so FiscalNote is a legal or RecTech company if you want to summarize it quickly.

David Joy:

Yeah. I mean, it’s really awesome because when I was researching after I spoke to you about FiscalNote, I was like, “This is a very, very interesting use case.” So what I did was I went and checked out Tim Hwang, who’s the co-founder, CEO. I went and checked out his profile a little bit just to see what kind of people you hang out with. So it was such a fascinating story of him. I think he was talking about how he was part of the Obama campaign and worked there and was really interested in making a difference and wanted to focus on what can we do around these laws and things that come out, and how can we make informed decisions based on that. And that was the introduction to me learning more about FiscalNote.

And what you guys have built here is a completely unique product and a company that focuses on this particular aspect. And it seems like you guys worked on this AI and built that from the very beginning, data and AI, it was core to what you were trying to do. And of course, you brought up natural language processing. I mean, last two years is basically a house party for GenAI and natural language… How were you able to foresee this is where it’s going? And early, when you joined the company, what were the core conversations you were having?

Vlad Eidelman:

For me, language was one of the frontiers that was just really obvious that the way that we interact with people should be more like the way we interact with computers. So there’s this, I would say, blip in human history that we went from no technology… Or no computers. Sorry. Obviously, there’s always been technology. The specific kind of technology, we’re talking about. Computers, and the way we interact with them and we program them, and then we’ve got keyboards, and mouses, and monitors, and the way that we program, and you have the special group of people who are able to write code in the way that computers are able to interpret it. And even that’s evolved, right? You went from assembly-level code to high-level abstractions like level one, two, three, four languages. And now you’re able to have people who are reasonably confident in programming computers without a lot of the fundamentals of understanding how all the hardware works. But now, you’ve got even further out to that point of not even knowing or not needing to know LLMs.

And so I was always very passionate about language as a way of interaction of transmitting information and as a way that we basically experienced the world and learned the world. And so I was taking that, but I would say that was also equally passionate about the way that we can create systems that learn. And so language just happens to be one of the things that we learn. We also learn vision and touch, and we can build robotics in the real world, but learning systems in general, that’s where my passion was. And so the evolution of those things just keeps going, the kinds of problems we solve and the challenges that we run into. And so when I was getting into the field, the biggest challenges were building these machine learning algorithms that were able to start generalizing from one domain to the next to be able to, again, scale it, to take a bigger data set than you can have on your one machine and be able to actually scale it across a number of different computers.

And so things like MapReduce were coming out of Yahoo, and Hadoop, and other places where you can start to implement that. Spark was just coming out and being able to parallelize different processing things. And so you had to build your own… It was already in the cloud. We didn’t start off building in FiscalNote on-prem, but it was early days in the cloud. And so we started off with Rackspace where we had to do a lot more of our own maintenance, and then we moved to AWS. And so the early-days conversation’s particularly around… Let’s go back to data and AIs. We knew that a lot of information was out there on the web publicly, but it was very hard to access. And so that’s one of the big, I think, challenges in general for companies that wanted to collect data from public sources is that, yes, it’s accessible, yes, it’s technically public, it’s not a proprietary data set, but it’s actually very difficult to access a lot of data sets that are public even on the internet.

And so if the governments were better at… Spent a lot more time putting out data sets in RSS feeds and APIs and keeping them updated and having databases that people could be accessing, then part of what a lot of companies like ours do would not be as useful. But given that data is so spread out across the internet, just saying, “We have data,” doesn’t mean much. It’s about how useful is it. It’s how useful is it to not just for you to be able to do something with, but then for you to present that to customers so they know how to deal with that and work on it. And so part of our moat and our differentiation from the very beginning was how can we get really good at identifying sources of data online, the public data sources that we think will be valuable to our clients. Then how do we bring those in efficiently. And so the sites break, there’s all sorts of things that change, things come up. Webmasters don’t like you scraping their site more than X amount of times per day.

We want to be good citizens of the internet, and so how do we make sure that we can build systems that are resilient to changes while also minimizing the maintenance time that we have on our side to be able to actually update and change. And then once you bring the data in, then how do you actually start to work on it? So you want to normalize it. We’re dealing with tens of thousands of different localities across 15,000-so cities in the US to dozens of countries and all of their different localities in different [inaudible 00:10:20]. And so we want to standardize it in a way so that when you ask a question like, “What’s my regulatory exposure to introducing a healthcare product in Brazil,” we want to be able to answer that question by bringing to bear a number of different sources. You and I are not thinking about, “Oh, well, you collected this information from Congress. You collect this information from Twitter.” You’re thinking about, “What’s all of the things that anyone has said written on that analysis for healthcare products in Brazil?”

And if we can bring that all together under the hood in our ways in that we process it and combine it in a way, and then expose it in a way that makes it valuable to you, that’s what we’ve been talking about from the very beginning. So on the outside, I think we are a legal or RecTech company. I think for users, we are trying to bring them this issues management, policy analysis, geopolitical market-intelligent platform. Under the hood, in the R&D department, we’re essentially an information retrieval and data product. And so we need to get really good at bringing in the right data, analyzing it, normalizing it the right way, and then marrying it together across each other so that the value of the data actually compounds the more of it we have and the way that we bring it together to then expose it in places that the actual user might want to use it, whether it’s searching over it, getting a [inaudible 00:11:29] on it, or then reporting or taking on some other action.

David Joy:

What you guys are doing and serving is serving data in a much more structured way from internet that is so unstructured and making it tangible for people to make decisions on or to understand what it is going to do. That was my read too. But it’s interesting what you were saying, right? You look back at 2010 to 2020, we will talk about four things that you just mentioned. The idea of MapReduce and Spark, and the birth of, in many ways… I don’t know if you remember the AlexNet paper came out in 2012 and it’s spurred a bunch of us… I was working at a company. And when the paper came out, I was like, “Oh, damn. This looks really fascinating,” but just I had no idea as to what this will evolve into until, by 2017, you had the transformation architecture come out, the attention is all you need and it just changed everything.

So it’s just been a decade of this happening. And I remember running my regression models and looking at my gradient boost and try to figure things out like, “What’s going on here?” Like, “It just didn’t make sense.” And I don’t know if you recall, at the time, I was programming on R, actually, which I hated really, until I started shifting to Python. So that was an interesting era. And I don’t even know if in 2012, a PySpark as a client was available for Spark. What were you guys using?

Vlad Eidelman:

I don’t believe so, but don’t quote me on that. So, I mean, this is funny where it shows you the mentality of where we were at a given time. So when I started at FiscalNote, I just assumed I should be building things from scratch. And so I started to rebuild the TF-IDF calculations from scratch for our search and for [inaudible 00:13:21] and max entropy classifiers, building them myself. And then obviously, I quickly found, at the time, I think, and still this package is sklearn and GenSim for [inaudible 00:13:33] modeling, and they had done a lot of the work standardizing some of the core APIs that you would want to make it easier, but certainly things like AutoML structures and being able to search the parameter space and all of that hadn’t come out. So we basically built our own versions of poor man’s AutoML, poor man’s MapReduce, versions of things that became more standardized and more available.

Because I think there’s also a difference in technology. I think there’s also a relevant… To today’s world of what’s possible. Maybe there’s a company that is coming out to do this and how infiltrated and how useful it is across the actual industry, the adoption curve basically. And so, yeah, there’s probably packages people are putting out on various things, but the likelihood that you’re keeping up with all the packages come out and that they meet your requirements for everything you want to do and that you can adopt them is probably pretty low. So there’s a lot of homegrown solutions. Again, at the early 2010s, I think a lot of people were building their own solutions at the time, and even though the open-source community was incredibly strong with other kinds of products, that set of machine learning and AI-enabled open-source packages is nowhere near whether it’s today. I mean, it’s night and day.

And again, what I was saying in terms of just opportunities for careers, I think, at that point, I would still argue you needed a PhD to be able to go into a product and be able to build out a lot of what you needed to. It didn’t have to be a PhD necessarily in computer science. A lot of folks were coming in through other physics, math PhDs, but you still needed some heavy understanding of the mathematical underpinnings of the actual algorithms to be able to provide and actually apply them. Whereas now, whether that’s good or bad, I guess we can also discuss, you don’t need a lot of that. You can come in and just jumpstart your abilities to build AI products without even understanding any of the underlying things. So great for some things because it just expands the number of people who can build products, and the value of that is incredibly high. Bad because when you’re running into problems, there’s fewer people who actually understand the core of what’s going on underneath to be able to really triage and solve the problem in the right way.

David Joy:

No, I will quote, from what you were saying, this idea that we have added a level of abstraction over the programmer and the language of communication is basically natural language. This goes back to… I don’t know if you saw Jensen Huang from NVIDIA, about a few weeks ago, said that everyone on the planet should be a programmer, that’s the goal, and that’s what everybody’s building. I’m really interested in what you’re doing with GenAI. Maybe we’ll touch that in a bit. But I’m curious, when you started FiscalNote, obviously you’re coming in, 10-people company, trying to build something that makes sense, create [inaudible 00:16:14] also evaluate if there is an opportunity for this and very early stages, right? And you had to quickly pivot from just being a scientist to looking at a product. And I’m pretty sure looking at it as a product owner at times, and seeing how to build product out, how to lay infrastructure out, how to think about scale, how did that transition happen for you?

Vlad Eidelman:

I think that’s one of the most exciting things about being in a startup anywhere, whether it’s successful or not, is the constant pace of needing to put on new hats and take off old hats and be able to evolve quickly. And so I think that’s what people talk about is maybe some people are not as decided by that and that’s totally fine, but if you’re excited by the drive that’s created by this resource-constrained timeline straight environment, I mean, there’s no better feeling I think in the world. My analogy is if you’re imagining you’re on a college project, except every member of that team is giving 110%, it’s like you’re not the only guy working on it, which is usually the case, and it’s four people playing something in the background. Everyone is dedicated 110% to that mission. It is incredibly powerful to see what that unified mission and vision alignment can do and to ignore some of the things that are not as critical to the day-to-day success.

And so while it’s absolutely necessary to have longer term vision in mind, you want to be able to have a roadmap in your mind of like, “Well, here’s where we think we can get to. Here’s our vision of what the world could look like.” We need to have that. If we’re successful in 10 years, we think the world will have a much more natural way of interacting with legal and regulatory data so that anyone can find what they’re looking for with as easy as a question. That could be a vision. The mission then is how do we apply that in our overall values, our structures day to day, what processes do we build to be able to actually implement so that people, as they come into the company, they’re quickly acclimated to the vision and understand how they fit in and what they can do and what the impact is going to be. And so every individual is empowered to make a decision day to day of what the most important thing is, what the least important thing is.

And that gets incredibly difficult. And I’m not saying that we’ve succeeded in doing that as best as we could, but I think that’s a challenge every company faces is basically creating that environment to empower individuals and early on… To get back to the root of your question, early on, it is very clear, I think, because it’s such a small type group, what is the most critical thing to focus on? So while scaling could be something that we know is going to happen at some point later, we’re going to hit that problem, that’s not the problem today. And so the first problem is making sure, like you said, that we’re actually building something that’s going to be useful for someone and that they’re going to want to pay for, because a lot of people will say, “Hey, this is really cool. I want to use this,” and then you ask, “Well, how much would you be willing to pay for it?” And then they’ll say, “Well, I don’t want to pay for it.”

And so depending on what your go-to-market strategy is, that might be fine, maybe you’re funding it through a different strategy where the user themselves is not paying for it, but someone’s paying for it somewhere. And so you have to figure that out at some point. And so I think that even if you have a lot of capital and a lot of investment and your focus is just growth and accelerating user growth, having some idea of what you’re going to do to prioritize going-to-market strategy, whatever it is, is super critical. So I think we were good at that, but I think we were good at having that bright trade off where we focused on short-term goals to hit milestones that we thought were really aggressive and we often missed, but at least we were overshooting where we needed to be, maybe not getting to where we want it to be, and then very critically prioritizing things that while cool, and useful, and valuable, and necessary like scaling considerations, certain processes, we’re just not critical to that mission of getting to where we want it to be.

And that’s painful too because then there are things that you look back on and say, “Six months ago, if we had implemented a better engineering ladder and having the visibility for the engineering team to know how they’re going to progress, it would’ve made it easier for this promotional cycle for people to understand where they fit in and what skills they’re going to have.” But that was painful, but we didn’t do that because we spent time doing some other consolidation that was actually necessary. So it is constantly a challenge in every role, but especially in the manager role to basically evaluate the important versus the urgent. So some things are urgent, you need to do that right away while other things are important. So you can’t always trade off the important for the urgent, but you can’t do the converse either.

David Joy:

In your answer, I saw your board of directors hat coming in as well, because you’re talking about looking at the strategy is the product viable and those ideas well, which I’m also curious about. But one of the things that I’ve really enjoyed working at startups, and I’ve, last seven, eight years, been working at startup series E, series F, that’s my story, especially moving in the last eight years. And what I’ve learned about, similar to what you said, is the feedback loop is insanely good. You build something, you talk to somebody, “Hey, I’ve built this,” the feedback loop is immediate here, “Hey, this doesn’t work,” or, “This is exactly what I need,” and that helps me understand how to iterate over the product. And the overall product just matures faster. And obviously you also have to figure out how many people can you go to and get this feedback from, because if that chunk is really small, then the potential for the company is questionable.

But it’s great to know that you’ve done that, you’ve thought about those ideas. So let’s get into a little bit more about… Because I’m always curious about you being a computer scientist and more on the data and AI side of things. How was the experience of the engineering side of things and were you involved on that or were you also involved on infrastructure side of things? Let’s talk about the recent years, how has it changed at FiscalNote?

Vlad Eidelman:

When I came in, I was an individual contributor on that team, and then as we grew, I talked about hats [inaudible 00:22:10] in and out, one of the hats that I started to give up more was the individual contributor hat and put on more was the manager and executive hat, which allowed me to focus both down toward what we need to do in our overall R&D strategy, what we’re working on, how we’re working on, but increasingly more on laterally on the other peers I had in the other organizations like our sales and revenue organization, marketing, operations people, et cetera, to start to build out the broader understanding of what it means to be at this company, which all direct back to engineering. Think of great product development, great engineering as not just the technical skillset for building and being able to code, but understanding why we’re here, what is it that we’re actually doing that might not be the core job.

So you might have a product manager who’s really going out there talking to clients, getting the feedback, validating the hypotheses and bringing that back, this is why we’re building it, but a great engineer should be asking those same hard questions from the product manager to make sure that as we’re prioritizing this limited valuable time of being able to either develop feature A or feature B or reduce the risk of some tech debt C, we are really putting our bets in the right places. As we’ve grown, we had one product that we had to do that for and instead of product managers who did that, and then we had basically three products over the course of the next couple years, and then we had 12 products and 15 products. And so we basically acquired over a dozen companies over the last couple years.

And so as they came in, there were different theories and thesis of why each of those products were a good addition. Some of them were really a partner to a product we had and we thought we’d expand on the market. Some of them were accelerating our coverage of the market, so maybe the same product, but creating synergies between the users that we can then migrate from the one to the other, and then sunset one of them. And so we had this larger set of customers, revenue-based, the same products that they were serving underneath, some of them were just a completely different market opportunity that we wanted to test and validate against whether it made sense for us to be in there. All of those came together in different ways on the engineering side.

And so over time, this is one of the things that I’ve spent my time on personally over the last couple years and how I’ve been involved, is creating the right organizational structure for us to be able to operate together instead of in these silos where we started off into creating these units of being able to have these squads or other cross-pollination to having a much broader R&D culture organization global values and initiatives where you can be on a horizontal platform team underneath and serving DevOps, SecOps, AI, analytic, or you can be on a product-specific team. But either way, you have an understanding of why we’re doing things, where we fit in, and how each of these is interacting in the longer-term vision. And again, it’s evolving so quickly, even at our stage, that sometimes it’s hard to keep up and there’s a lot of communication, and that’s almost always undervalued. No matter how many times I say communications is key, it’s not saying it enough, basically, communicating all of these things.

David Joy:

You brought up customer acquisition. And I have seen from my experience that there are companies who have done that really well, and there are companies who have done that really badly. And one of the things I’ve observed is that, as a challenge when you do customer acquisition and you have a similar product or a product that you want to integrate with, is that they’re on a completely different stack. I believe you guys are in AWS, that’s what you were talking about. So did you face any challenges with having a product that you acquired that was running on a different stack? How did you go about thinking about how should we bring it in or did you custom bring everything in, just move containers over multi-cloud? What was the strategy there?

Vlad Eidelman:

Yeah, you’re absolutely right. We’ve had products that are running on-prem in a data center, running in a different cloud provider, running in AWS, but very differently from any of the ways that we’re running with different CI/CD tools and different infrastructure, different code repositories in ways that they’re doing their software development life cycle. And so there’s a formulaic corporate answer, and then there’s the real-world answer, right? The corporate answer is that-

David Joy:

Yeah, give me the corporate one, and then the real world.

Vlad Eidelman:

Yeah, I’ll give you the quick corporate one, which is everyone’s going to say they have a playbook. So do we. So when we acquire a company, pre-acquisition, during discussions and diligence, we’ll identify the product and technical security, the places that we’re at. We’re just taking a temperature, right? It’s not like there’s a wrong or right answer. It’s very rare that that part of the discussion is going to torpedo the acquisition thesis. Obviously, if there’s huge red flags and if we’re acquiring it for technology, that might. But in most cases, we’re looking at it from a much broader perspective of the opportunity of the market, the people that are at the company, and just the synergy for what they can bring. And so the tech product is just making sure that we have a reasonable understanding of what is it going to take for us to operate this? Is it something that’s going to take a huge effort where we’re going to have to divert a lot of resources to get it to a standard that we want or is it at a place where we can reasonably keep it where we are?

And so that’s part of the first piece. The second is once we’ve basically made the acquisition, there’s a playbook for [inaudible 00:27:23] things that have to happen and things that could happen. So things that have to happen, we have to have access to the code repository, source code, the actual infrastructure that’s running on like admin access to all the things so we can run our security-based level of tooling or penetration tests, other things that we need to know and run like our code security gaps and things like that. So that meets our SOC requirements and other things. We have to have those things that happens first, and that’s where the DevSecOps team comes in and they work with whatever the organization’s leads are to do that.

Then the things that may happen, and this is where it depends, is the practical reality of does it make sense for us, the ROI of doing this work, the investment versus other things. So just because you’re using a different database, I mean we have Postgres, we have MySQL, we have Oracle, and we probably have some others. I know we have some others. So does it make sense for us to move you from Oracle to MySQL or Postgres just because most of the org is running Postgres? Probably not. Are there other factors involved in that decision? No. Are we going to migrate that data set in some other data lake infrastructure that we don’t need to even know what the database underneath of it is? Is it costing us? Is there a license to that database that we have to pay in that as opposed to improving it somewhere else? So we make that decision on a case-by-case basis, and so some things will leave as they are because just doesn’t have an ROI to migrate.

Other things, we do have a plan for as part of a broader strategy. So authentication and SSO. We want to be able to get to a place where, across any suite of products, it’s seamless to be able to integrate an access from a user perspective and from an internal coverage perspective. And so that we will invest in, and that becomes part of the need-to-have playbook as we evolve that and that we have a way of doing that, that we know exactly what we’re looking for from each team. So as we’ve done this over time, the need-to-have part of the playbook has gotten bigger, and we know how to do it, and then there’s still the case-by-case decisions on any of them that are going to be which way do we go is basically a trade-off conversation.

David Joy:

That’s probably the most quality approach, because you don’t really know until you see the source code and what the realities are that you have to debunk or unbunk, right? So one of the other things that I was curious about was that, right now, is it your product as FiscalNote products are all B2B or B2C customer? Is B2C?

Vlad Eidelman:

Primarily, it’s B2B. Primarily, our clients are organizations that are subscribing to our products.

David Joy:

So I’m curious, why not have an open product for B2B users? Say, I’m just curious. I’m as a user saying, “I want to know a legislation came out, how does that impact me and place I live in?” Do you think that is something… I’m just curiously throwing an idea out over there. So, yeah.

Vlad Eidelman:

Of course. So we have a couple in front of the paywall properties where we have content like rollcall.com, factbase.com. There’s things that you could see on Oxford Analytica or FrontierView, if you’re not a subscriber, that we still want to create more information out there for more accessibility. From time to time, we have launched public-accessible projects. You can see COVID updates… When COVID launched, we had, as everyone did, a COVID tracker center and some other things about legislative officials and tracking things. And so I think coming back from a strategic perspective, is it something that we evaluate? What is our differentiated advantage? And I think that’s important for every company. Is it the actual information? Is it how we process it? Is it the accessibility of it?

I think there is definitely opportunity. In the last couple of years, product-led growth has been much, much bigger part of, I think, companies, and a lot more startups are coming out with a free tier or an early small paid tier, especially AI companies that have a copilot [inaudible 00:31:04] strategy. You have a $10 per month per user. And then if you see that there’s enough of that, then basically the inside sales teams comes in and says, “Hey, we see that there’s 100 people using this product, would you want to get an enterprise contract?” So I think that’s part of the strategy we’re evaluating where that might make sense. And so, absolutely, it’s something that we think we want to do. And then there’s also the public civic good, which is what you’re talking about. Can we also make it easier for people to be able to understand their own government? And so that’s something that we’ve also had some inklings of doing, and then I think we might do more.

David Joy:

Yeah. I’m curious, actually, when you guys started working in FiscalNote, was there any other existing competitor at the time that did something or was trying to build something similar or where you guys were the-

Vlad Eidelman:

Sure, of course. I mean, I think one of the first hires that we’ve made was head of sales at a number of really large recognizable consumer companies. And one of the things I took away from him was he came in [inaudible 00:31:58] it was early stages at the company is that now everyone would know, and he said, “When we started, we weren’t sure we had a product until this other company and this other company came along and started to try to do the exact same thing.” And then we knew that there was a validated product need, because otherwise we basically have a solution, but we’re not sure if anyone recognized the problem. If you have a couple other people in the space that helps validate that this is in fact a real problem, that there’s an opportunity, and it should be a red flag if you’re the only ones doing something. I think that it’s actually really healthy to have competition both from the incumbents and in the startup space.

And so now we’re in the middle where we’ve been around, we’re public company, but we’re still small. And so relatively, on the bigger side, there’s information services companies like LexisNexis, Wolters Kluwer, Bloomberg, Thomson Reuters, a lot of them have offerings that are very close in some ways in legal and policy and political information from ours. On the other hand, there’s the economist group in other geopolitical analysis firms. On the other hand, there’s other advocacy tools. Because we covered such a broad area, there’s a different set of competitors in each of these markets, and there’s some that are closer to align to some of our products than others. But, I mean, I agree that part of the counter to what I was saying, but it’s still part of it is you have these competitive spaces, but you want to try to create your own niche.

And so we’ve tried to basically carve out a space where we feel like part of our advantage is this end-to-end connection of these different aspects of the products, whether it’s at the data layer, some of the analysis services or the application layer. I mean, all of those are competitive potentially advantages that we can say that there’s a synergy in some of these things that you don’t have access to if you only have some of these other competitive products.

David Joy:

You guys have been doing this for a decade, obviously, but with the… I’m really excited about the amount of new companies that are going to get started and fail or remain because of all the… A little bit of a FOMOB around AI, but you guys have vetted this AI process for a while. That’s your bread and butter. So you really know the advantage of how you’re going to use GenAI in your use cases because, of course, there’s summarization is one of the things that we see at Q&A and stuff like that, but the opportunity to fine-tune your stuff, store that in a RAG model and serve it to your users is a great opportunity. So anyways, I was curious about what are you guys doing with that.

Vlad Eidelman:

As everyone, we’re still figuring out exactly where it provides value. Internally, we have a lot of subject matter expertise where we have people who are analysts in certain countries in certain industries, and they’ve been creating geopolitical analysis for France or Nigeria or Germany or the energy sector for decades. And so we have this huge database going back 50, 60, 70 years depending on the product of all of this analysis, really thoughtful structured data and unstructured analysis with charts, and graphs, and databases of different indicators and risk indicators, et cetera.

And so there’s so much structured and unstructured data that we can tap into, and then we have this public data side that we’re basically ingesting a lot of this information and storing it. And so there is an advantage in some of that disappears from the public internet, and yet we still have that archive. There’s some of that, but I think a lot of it is very timely. You want to know certain things about news now or social media now. And so I think we have both of these products and both of them have internal people that work on them. There’s engineers working on the data ingestion side and refinement. There’s the subject matter expertise and analysts working on understanding and getting the right data source and content. And so LLMs and generative AI have a transformative power in both of those internal operations. Basically transforming work that could only be done by an engineer to a non-engineer is a huge efficiency gain, because then you can really spend your time in the most optimal ways, and then encourages leveling up on both sides.

And then there’s the actual content creation work. So the editorial, the journalism, the analysts, and actually synthesizing information. So that’s where the power of that has been, where there’s really simple things that companies struggle with. We basically have a Google Drive or whatever it is, a SharePoint, with thousands of files, like, “When did we talk about this? Where is this?” Everyone has that problem. And there’s startups that are literally just trying to tackle that problem. I mean, we have that problem too. So we have all of this data, but sometimes even our internal teams are trying to marry things that we have internally and aren’t sure where to find it or where it is. And so being able to provide that so they can create better, more informed content. But then obviously the harder part is how do we expose that to users? What are the products that we’re working on?

And so we have a number of tiers that we think about it as some of them are going to be much lighter-weight copilot opportunities where you don’t need a whole workflow, you don’t need a whole bench of different ways that you can work and interact. You really have a targeted question. You want to come in similar to maybe an expert network and say, “Hey, can I have an analysis of this so that I could put into an email a PDF, show it to my boss,” whatever, “All the way through, I want to build an entire 50-page report on what it’s going to look like, what is our risk exposure if we launch a new product in the US,” or, “How do I respond to this new regulatory filing that’s going to ban the part or the service that I provide in New York?” And so there’s a lot of work, a lot of cognitive work and a lot of actual manual work that goes into doing these things.

And so we have a lot of data, we basically have been working on how to correctly embed it and be able to combine it in ways that make it useful across the different data sets that we’ve been able to both acquire externally and create internally, and then be able to have the right chains of agents, basically, if you’re thinking about an LLM to be able to confidently consistently answer questions. Because the big concern for us is people trust us. We have a reputation for providing actionable, insightful intelligence, and so what we don’t want to do is ruin that by creating hallucinations or misrepresenting the information we have, but we also know that there’s an expectation from customers that they’re going to start receiving a much faster turnaround on that and that there is going to be some error rate that people are willing to stomach as long as it’s reasonable.

So 80% of the work has been in making sure that the things that we’re providing are actually reasonable and don’t have customers lose trust in our ability to provide the right product, and then I think what we’re going to put out in the next month actually, or two, we’re launching a few things, is going to be really exciting because I think it’s exactly where the market needs to move.

David Joy:

Yeah, it was really profound answer. I mean, I love the fact that you have been thinking about it and… Because, I mean, I don’t know anybody who’s not. And obviously, add another vector database to your database portfolio for storing your embeddings, right? It’s interesting because I know Postgres has pgvector support, which supports embeddings, but there are other embedding databases. Our company have been also having conversations around should we provide that capability and we serve for what Cockroach does, serves a tier one, tier zero application that requires high availability, zero downtime, and that is our primary use cases and it’s largely OLTP transaction.

So again, going back to what you were saying, we have to look at where this demand really is or if it’s just something somebody wants to see as a feature but might not primarily use. So these conversations happen all the time. So now that you’re a CTO, and a decade ago, computer scientist working on code, doing stuff, do you get to dabble with code nowadays? Do you go in, run those RDDs on Spark or do some analysis or do you do some personal products for fun and figure things out?

Vlad Eidelman:

Yeah. One of the lessons that I learned probably too late, if you ask my team, is that I cannot commit to being part of a sprint or a combine or some process. It’s going to depend on me creating code for our software development. And so I stepped back from that a number of years ago. What I try to do is basically still play around with things off of the side, so it could be a research project. So we also try to do some things, again, given my academic background, that involves some amount of just going a little bit further than the product case to an academic question and being able to publish, and go to conferences and be a part of that community. And so I still try to play around with that and be able to understand what people are working on.

And I guess maybe tying into one of the other lessons is when you have the opportunity to create yourself as a manager and understand where your skillsets are, you are going to get to a place at some point that the actual teams, the functions that teams are performing are outside of your purview. And I think that’s actually good, right? I cannot step into the role and actually do the thing that every single person on my team does. I don’t know the frameworks on the application side. I don’t know the security tools on the DevSecOps side. And so it means that it forces me not to think about, “Oh, how would I do this,” or like, “What does this look like,” but to ask more questions that empower the person to make their own decision and just bring the context and understand things. And so I think even though I have a high passion for AI components in the data science, I would not impose myself on the team to be saying like, “I should develop this part,” because that would be disservice both to the project and to them in a lot of ways.

But I do like to maintain some amount of understanding so that I can ask reasonable questions and dig in at a sufficient level of the technical depth to be able to at least poke at the things that might be issues or that we might be wanting to pay attention to, that I might be either seeing through patterns about what other teams are working through or struggling through or externally. So I have a number of companies that I have pretty close relations to, and so I can basically pattern-match across them what people are running into and where problems are, and to bring that context back. I think that’s one of the biggest things that I could do as a valuable contribution to the team.

David Joy:

In your free time, how do you continue to learn about what’s happening in the space? Because every damn day, there’s some new stuff that comes out and you have to learn to filter stuff out. How are you managing all of that?

Vlad Eidelman:

Yeah. So I think, again, I don’t know that I would recommend getting a PhD to most folks now. It really depends on what you want to do. But one of the big values of getting a PhD is the psychological comfort that comes after a while. While you’re very uncomfortable for a long time, you feel like everyone is scooping you, you have ideas and you validate and you think, “Oh, this is a new idea,” and then you realize everyone has already done this or knows this, and then you go through these roller coasters, and you’re also working on something for years, right? It’s not like you can implement a feature and get feedback. It’s really like your thesis takes years.

And so you start to build patience, understanding. And I think that there’s a certain kind of appreciation that comes with that that you’re not going to be able to keep up with every single thing. And so what I try to do is I have a number of newsletters that come in from various sources. There’s the various threads on social media that I’ll follow or various posts on news articles. But honestly, the backstop is usually people that I know are pointing to something, and I think that if this is important enough and it’s something that is causing enough of a stir, I will hear about it from someplace. And if I don’t, that’s okay. Looking back, I’m sure if you did this experiment, look at what you read last year and how much of that was relevant and how much of that actually formed your opinions and didn’t, probably a very, very small amount.

Now the question is you never know which part of that small… Is it going to be like, “It’s the marketing [inaudible 00:43:29],” right? 50% of the marketing works, you just don’t know which 50%. But still, it should give you some comfort that you are able to not follow everything, and then you’ll probably hear about something that’s important, and it’s okay if you don’t hear about it the second it comes out, probably the next day this week. There’s something that will, I think, catch your attention.

David Joy:

Segues into my one question was I was going to ask you what is the one lesson you want young engineers, computer scientists or people who are on a similar journey as yours or exploring a future that looks like yours as a CTO in some… What would be your one lesson to them?

Vlad Eidelman:

The successful engineer at whatever level, starting from an SE1 all the way through staff or principal, knows how to work with both technical and non-technical people. And I think that, again, one of the things I was thinking about when I went to go with my PhD is, “Oh, good. I won’t have to spend as much time networking as if I went to work at a business right away or some other things.” And I quickly found out that’s not the case. You basically still have to spend as much time learning what the other students are working on, learning the professors, networking in conferences to understand where the research is, what people are doing to, like you said, get information. And so the ability to understand and work with people is basically the same as your ability to influence where the direction of the project is going, the company is going, and to guide your own development and skills.

And so I think the one lesson is to get really good at being empathetic, to understand where the other person’s coming from. And basically, if you’re working with a product manager to get the question of why and why we’re working on something and understanding the users, if you’re working with your other technical peers to learn from them what drives them, what motivations they have, it’s applicable in your personal life too. And so I think that that’s the one thing I see that needs more development in people, is that they get really good at some of the technical aspects and they dive in, but they neglect the need for understanding and working with other people. And that just really limits their effectiveness. And so if you want to be effective, if you want to feel impactful, I think the ability to communicate and work with other people is one that’s so obvious, but often so overlooked.

David Joy:

Yeah, brilliant. I mean, spoke like a true philosopher. I really wanted to get into the human philosophy side of things for everybody listening.

Vlad Eidelman:

Well, we just-

David Joy:

So, yeah, I know.

Vlad Eidelman:

Yeah.

David Joy:

But, I mean, I had read through some of the posts you had like your posts around a little bit of envy is okay for humans and machines. I did want to get into some of those ideas, but I have to be cognizant of the time. There you go. It’s been an absolute pleasure having you on, Vlad. And the way I see this is first of our conversations because I feel like there is so much interesting ideas that you have. For everyone listening in, if you want to follow Vlad, he’s on LinkedIn as well as you have a blog post called machineopinings.com, with the opining is O-P-I-N-I-N-G-S. Basically, there’s no E, if I’m not wrong. Opining. Opinings. Yeah, there you go.

Vlad Eidelman:

That’s right. Machine Opinings. Opining.

David Joy:

Thank you so much once again for coming. And thank you, everyone, for listening in. It’s been a great time talking to you, Vlad. So see you in the next one.

Vlad Eidelman:

Thank you, David. It’s been my pleasure.